matmul raises IndexError exception with input shapes (2, 1, 2) and (2,)

[antonwolfy]

The below example causes an issue:

import dpnp, numpy

dpnp.__version__
# Out: '0.17.0dev4+3.g498e705d848.dirty'

a = dpnp.ones((2, 1, 2))
b = dpnp.ones((2,))

a @ b
---------------------------------------------------------------------------
IndexError                                Traceback (most recent call last)
Cell In[10], line 1
----> 1 a @ b

File ~/code/dpnp/dpnp/dpnp_array.py:489, in dpnp_array.__matmul__(self, other)
    487 def __matmul__(self, other):
    488     """Return ``self@value``."""
--> 489     return dpnp.matmul(self, other)

File ~/code/dpnp/dpnp/dpnp_iface_linearalgebra.py:851, in matmul(x1, x2, out, casting, order, dtype, subok, signature, axes, axis)
    846 if axis is not None:
    847     raise NotImplementedError(
    848         "axis keyword argument is only supported by its default value."
    849     )
--> 851 return dpnp_matmul(
    852     x1,
    853     x2,
    854     out=out,
    855     casting=casting,
    856     order=order,
    857     dtype=dtype,
    858     axes=axes,
    859 )

File ~/code/dpnp/dpnp/dpnp_utils/dpnp_utils_linearalgebra.py:951, in dpnp_matmul(x1, x2, out, casting, order, dtype, axes)
    949         else:  # call_flag == "gemm_batch"
    950             assert call_flag == "gemm_batch"
--> 951             result = _gemm_batch_matmul(
    952                 exec_q,
    953                 x1,
    954                 x2,
    955                 result,
    956             )
    958 if NumPy_special_behavior:
    959     result = dpnp.tile(result, out.shape)

File ~/code/dpnp/dpnp/dpnp_utils/dpnp_utils_linearalgebra.py:373, in _gemm_batch_matmul(exec_q, x1, x2, res)
    371 x2_shape = x2.shape
    372 x1 = dpnp.reshape(x1, (-1, x1_shape[-2], x1_shape[-1]))
--> 373 x2 = dpnp.reshape(x2, (-1, x2_shape[-2], x2_shape[-1]))
    374 orig_shape = res.shape
    375 res = dpnp.reshape(res, (-1, orig_shape[-2], orig_shape[-1]))

IndexError: tuple index out of range

# works with numpy:
na, nb = a.asnumpy(), b.asnumpy()
na @ nb
# Out:
# array([[2.],
#        [2.]])